"We Want a Semiconductor Industry in Taiwan"

In late August 1985, a fifty-four-year-old Chinese-American executive in a dark suit walked through a glass door at Hsinchu, on the northwest plain of Taiwan, and into a meeting that, by his own later account, he had been resisting for four years. The building was the headquarters of the Industrial Technology Research Institute, ITRI, a state-funded laboratory complex that the Taiwanese government had set up in 1973 to do for the island what the Massachusetts Institute of Technology and the Bell Labs of an earlier era had done for the United States. The man’s name was Chang Chung-mou, romanized for the American business cards he had carried for thirty years as Morris Chang. Until two months earlier, he had been the president and chief operating officer of General Instrument in New York. Until two years before that, he had been a group vice president at Texas Instruments in Dallas, where he had spent twenty-five years and where, the previous decade, he had run the company’s worldwide semiconductor division. He had a PhD from Stanford, a wife, an apartment in midtown Manhattan, and no particular reason to be in Hsinchu except that the Taiwanese premier had asked him to come, and the men around the premier had not stopped asking. He had agreed, finally, to take the presidency of ITRI. He understood, by the time he arrived, that the presidency was the smaller part of what was being asked of him.

The man who had asked was waiting in an adjacent office. Li Kwoh-ting was seventy-five, had earned a physics degree at Cambridge under Ernest Rutherford, had abandoned his Cavendish doctoral work in 1937 to help his country fight the Japanese, and had spent the next half century doing in the Republic of China what physicists almost never get to do: he had run economic ministries. He had been minister of economic affairs, then minister of finance, then a minister without portfolio with a science-and-technology brief. The American press, when it noticed him at all, called him K.T. Li. The Taiwanese press by 1985 was already calling him the Father of the Economic Miracle, and treating his recommendations as a kind of secular fatwa. He was the official the cabinet listened to when it wanted to know what the next decade should look like.

Li had been working on Chang for years. The two had first met in the late 1960s, when Texas Instruments, looking to put cheap assembly labor next to its Dallas wafer plants, had built one of its earliest offshore facilities outside Taipei. Chang had flown over to negotiate. Li, then in his economic-ministry phase, had been the Taiwanese counterpart. Chang would later say of those early visits that Li impressed him not the way a politician usually impresses an executive, with charm or veiled threats, but in the way a peer scientist did, by asking better questions than expected. They had stayed in loose contact through the 1970s, exchanging Christmas cards and occasional phone calls. In 1981, after the death of Chang’s longtime mentor at TI, the founder Patrick Haggerty, Li had written to Chang and invited him to Taiwan for a visit. The visit had produced a polite report and an unsubtle suggestion that Chang come to run ITRI. Chang, still a TI lifer at that point, had said no. In 1982, Premier Sun Yun-suan, a hydraulic engineer turned politician who had been the man behind the 1974 founding of ITRI itself, had called Chang directly and made the offer again, this time with terms. Chang had again said no, gracefully but firmly. He could not, he later wrote, work out a leave arrangement with his boss in Dallas; in private, he understood that he was at TI for life and that the work he wanted to do was in semiconductors, not in running a government laboratory in a country he had not lived in.

The fragility of that conviction had then been demonstrated to him, in stages, over the next three years. In 1983, after a quarter century at TI, Chang had been passed over for the chief executive’s chair. The board had picked Jerry Junkins; Chang had been moved sideways, into a quality-and-productivity role that he understood to be an antechamber to retirement. He had resigned that summer, accepted the presidency of General Instrument, discovered within twelve months that the conglomerate’s leveraged-buyout culture made him miserable, and resigned again in the spring of 1985. He was, when Li and Sun came back a third time that year, between jobs in a way he had never been in his adult life, with no obvious next chapter in American semiconductors and a growing suspicion that the chapter that was on offer in Taiwan, however unfamiliar, was the only one being written for him. He took it.

What Li had not yet told Chang, on the day Chang flew into Hsinchu to assume the ITRI presidency, was the second half of the assignment. By Chang’s later account, given in oral histories and in two volumes of Chinese-language autobiography, Li waited about three weeks. Then, in a meeting that has acquired in the retelling a rough granularity rather than a precise transcript, Li said something close to this: ITRI is a laboratory. We do not need only laboratories. We want a semiconductor industry in Taiwan. We want you to start the company.

Chang, by then sixty days into a job he had not particularly wanted, asked what kind of company. Li said he did not know. He said the government would put up about half the money, that he could find Taiwanese private investors for some portion of the rest, and that he expected Chang to find a foreign technology partner for the remainder. He said the government wanted a chip company that could compete on the leading edge with the Japanese, and that the chip company was Chang’s to define. The amount of money on offer, in absolute terms, was modest by American standards. The amount of latitude was unusual.

The latitude mattered because the model Chang had in mind by 1985 was a model that no investor in his right mind would have approved on a structured pitch. Chang had been thinking about it, intermittently, since 1976. In that year, while still at Texas Instruments, he had drawn up a strategy paper for the company’s internal planning process arguing that the cost of a state-of-the-art wafer fab was rising on a curve that would, by sometime in the 1980s, exceed the budgets of all but the largest integrated semiconductor companies. The chip industry’s response to that pressure, he had argued, would be a separation: design houses would shed their fabs and concentrate on circuit design; a smaller number of foundries would absorb their manufacturing and run the fabs at the brutal volumes such investments demanded. He had imagined, in the paper, that TI itself might play the foundry role. The paper had not gone anywhere. TI, like every other vertically integrated chipmaker of the era, considered manufacturing inseparable from design. The strategy paper had gone into the file. By 1985, Chang had quietly nursed the idea for almost a decade, watched the costs of fabs continue to climb, watched a handful of younger entrepreneurs in California talk about going fabless, and decided that the right answer was not a sometime-foundry that did manufacturing on the side. The right answer was a pure-play foundry, a company whose only product was wafers, whose only customers were other chip companies, and whose only competitive promise was that it would never compete with them. It would design nothing of its own. It would brand nothing. The customers would own the chips. The foundry would own the line.

This was, in 1985, an almost embarrassingly heretical proposition. Every reputable chip company in the world was vertically integrated. Intel, Texas Instruments, Motorola, AMD, NEC, Toshiba, Hitachi, Fujitsu, Mitsubishi, Philips, Siemens, SGS, Thomson — all of them designed and manufactured their own products under their own roofs, and all of them treated manufacturing yield as the proprietary alchemy that protected their margins. The conventional industry view, repeated to Chang in some version by every executive he later approached for funding, was that the people who knew how to run a fab also knew how to design a chip, that the two competencies fed each other, and that a fab without a captive design team would be condemned to subscale runs of weak product designs from second-tier customers. A pure-play foundry, in the standard 1985 view, would be a fab that nobody bought from. The market it claimed to serve did not yet exist at any scale that would pay for the fab.

Chang, who had spent twenty-five years actually running fabs, believed this was wrong on a structural level the executives reciting it had not thought through. In the United States he had watched a generation of younger engineers leave Fairchild and TI and Motorola to start chip companies of their own, and he had watched almost all of them spend the first half of their venture money building fabs they did not really know how to fill. Each of them was reinventing manufacturing inside a startup, badly. Each of them needed enough captive volume to amortize a fab against, and very few of them had enough captive volume, and so each of them was in a slow race to either grow the volume fast enough to keep the fab alive or sell the company before the fab killed them. A foundry, properly run, would let a chip designer skip the entire problem. A small team with a good circuit could go from idea to silicon without raising the hundred million dollars a fab cost in 1985. The sluice gate that had kept innovation locked behind capital would open. The market that did not yet exist, Chang believed, was a market that would create itself the moment the foundry did.

It was a bet on a feedback loop. Whether the loop would close fast enough to keep a real company solvent in the meantime was a different question. Chang had no answer to it that he could prove. He had only the conviction, half from his TI strategy paper and half from his California observations, that Taiwan was the right place to make the bet because Taiwan was the only place where the bet could be made cheaply enough not to ruin the bettor. Taiwanese engineering wages were a fraction of American wages. The ITRI laboratories had spent the previous decade absorbing seven-micron CMOS technology from RCA in a transfer Pan Wen-yuan, the New Jersey-based RCA scientist of Taiwanese origin, had brokered in 1976; ITRI had taken that technology to four-and-a-half micron by 1985 and had a working pilot fab and several hundred trained process engineers. Hsinchu Science Park, opened in 1980 and a personal project of Li’s, sat next door to ITRI with land and tax holidays and undergraduate-trained electrical engineers from Chiao Tung University and Tsing Hua University walking in every September. The pieces were lying around. Whether Chang could pick them up and assemble them into a company whose business model nobody believed in was the question Li had handed him.

Chang spent most of the autumn of 1985 and the spring of 1986 raising money he could not raise. The Taiwanese government had committed, through the Executive Yuan’s Development Fund, to subscribe what would eventually settle as 48.3 percent of the company’s initial equity. Li had then made clear, in his quiet way, that the rest had to come from somewhere outside the government, both to discipline the venture and to give it credibility with foreign customers. Chang flew to the United States and Europe and made the rounds. He went to Texas Instruments in Dallas and pitched his old colleagues. They declined. He went to Intel in Santa Clara. Intel, which under Andy Grove was at that moment exiting memory and concentrating on microprocessors, considered the foundry idea with mild interest and then declined. He went to Motorola, which declined. He went to Mitsubishi and to Sony in Japan. Both declined. He went to several European national champions. They declined. Each meeting was variations on the same conversation: the Taiwanese partner had political reasons for wanting the deal, the technology transfer was potentially useful, but the foundry concept was a thing nobody on the other side of the table believed could work, and the people doing the declining felt they were being polite by saying no. Chang’s later summary of the season was that the chip industry of the world had, in effect, voted against him. He had only one yes left to find.

He found it in Eindhoven. Philips, the Dutch electronics conglomerate, had been a Taiwanese industrial presence since 1966, when it had built a television-tube factory in Kaohsiung, and the relationship had since broadened into consumer electronics, lighting, and the small Taiwanese semiconductor operations that Philips ran under contract for Asian customers. Philips by the mid-1980s also held one of the world’s most valuable collections of semiconductor patents, including the LOCOS isolation process that nearly every CMOS fab in the world had quietly licensed and the cross-license web of mutual immunity agreements that the company had negotiated through the 1970s with IBM, with Intel, with Bell Labs, with Siemens. To start a chip company in 1986 without access to that web was, for practical purposes, to start a chip company that any large American or Japanese competitor could put out of business with a single patent suit. Philips’s seat at the patent table was the asset Chang most needed. Philips, for its part, wanted access to Taiwan in a deeper way than its existing Kaohsiung operations gave it; the deal Chang was offering was an entry point to the island’s manufacturing wage structure and to the local government’s continuing goodwill. After several rounds of negotiation in 1986, Philips agreed to put up 27.5 percent of the new company’s equity, contribute its production technology under license, and extend to the joint venture a sublicense to its third-party patent agreements, a sublicense that would give the new company structural air cover against most of the patent suits that would otherwise have been the first weapon used against it. The remainder of the equity, around 24 percent, would come from a hand-picked syndicate of Taiwanese industrial families: the Wangs of Formosa Plastics, who took 5 percent at the urging of President Chiang Ching-kuo’s chief secretary; the Cathay Group’s CAPCO; Taiyuan Textile; a handful of others; and, at 4 percent, the investment arm of the Kuomintang itself.

The new company was incorporated on February 21, 1987, under the name Taiwan Semiconductor Manufacturing Company. Its registered capital was about NT$1.3 billion, roughly forty-five million U.S. dollars at the time, with later subscriptions and lines of credit pushing total capitalization above a hundred and forty million. It was housed initially inside the ITRI complex at Hsinchu Science Park, where its first fab, a six-inch line running a 2-micron process licensed jointly from Philips and from ITRI’s RCA-derived stack, was carved out of an existing ITRI facility along with ninety-eight engineers transferred from the institute’s electronics division. The company’s first president and chief executive, on Chang’s recommendation, was an American: James E. Dykes, who had run the Philips North American semiconductor business and who Chang had recruited on the theory that a Western face would help with U.S. customer development. Chang himself was chairman, splitting his time with the ITRI presidency in the early months and gradually converting his portfolio entirely to TSMC over the following two years. Dykes lasted a year. He clashed with Chang over operational decisions and over how aggressively to push the foundry pitch on customers who were not buying. He resigned in 1988 and Chang, by then fifty-seven, took the chief executive role himself. He would hold it, with a brief retirement and a return, until 2018.

The company’s first eighteen months were, by every internal measure, terrible. The fab produced wafers, but the customers Chang had spent the previous year courting did not show up at the volumes he had promised the board. The pure-play foundry pitch played, in California and in Tokyo, exactly the way the venture meetings of 1986 had played: politely and to no purpose. The chipmakers who had captive fabs did not want to send work to a Taiwanese unknown. The fabless startups who would have been Chang’s natural customers were a thin handful, none of them yet large. The first real foundry orders came in the form of overflow from Philips itself and from a few mid-tier American firms that needed older-generation parts and could not justify dedicating their own fabs to them. The numbers in 1987 and the first half of 1988 hovered just above the line between catastrophe and survival. Chang spent the months he was not in California or Japan walking the fab floor in Hsinchu, talking to engineers, learning, at fifty-six, what running a Taiwanese line was like in detail. He had not personally been responsible for the inside of a fab in over fifteen years.

The thing that broke the company out, when it came, came from Santa Clara. By late 1988, Intel was running its 386 microprocessor at full volume on its U.S. and Israeli fabs and had a backlog on older 286 and embedded controller production it could not clear without diverting capacity from the 386. Andy Grove, the Intel CEO whose company had declined to invest in TSMC two years earlier, sent a delegation to Hsinchu to evaluate whether TSMC’s process was good enough to run a small slice of Intel’s older parts under a contract foundry arrangement. The delegation came back with a list of about two hundred process improvements the Taiwanese fab needed to implement before Intel would consider it. Chang accepted the list as written. TSMC’s engineers, working through 1989, executed roughly two hundred of the recommendations. Intel signed the contract in 1990. The order was small in absolute dollars and large in what it advertised. Intel did not award foundry contracts to companies it considered second-rate. The fact that TSMC was now running silicon for Intel meant, to every other potential customer, that TSMC’s process could be trusted. Within eighteen months, the company’s order book was full and its second fab was under construction.

The fabless companies that Chang had built the foundry to enable began arriving in numbers shortly thereafter. In Sunnyvale, a thirty-year-old engineer named Jen-Hsun Huang was sketching, on the back of a Denny’s placemat, the architecture of a graphics company he would call Nvidia. In San Diego, the founders of Qualcomm were imagining a wireless modem. Henry Nicholas and Henry Samueli, in Irvine, were drafting the first chip architectures for what would become Broadcom. None of these companies had a fab. None of them needed one, because Chang’s company in Hsinchu had announced, in effect, that fabs were optional. The fabless model that would by the mid-1990s become the dominant way new chip companies were built had needed two preconditions to take hold: a customer base of designers willing to outsource manufacturing, and a manufacturing partner trustworthy enough to accept the outsourcing. The first precondition was forming organically as a new generation of VLSI design tools lowered the barrier to chip design. The second precondition would have remained missing for years longer if Chang had not built TSMC. By the time the dot-com expansion hit at the end of the 1990s, the foundry pattern was the model. Companies that did not yet exist were already assuming TSMC would be there to manufacture their first chips.

Inside Taiwan, the meaning of what Chang had built took longer to register, in part because the country’s economic story in the late 1980s and early 1990s was so crowded with success stories that one more chip company hardly seemed remarkable. Acer’s Stan Shih was building a personal computer brand. Formosa Plastics was building plants in Texas. The Hsinchu Science Park around TSMC was filling up with smaller fabless startups, with mask makers, with packaging houses, with testing operations, gradually condensing into a complete semiconductor cluster of a kind that did not fully exist anywhere outside the United States and Japan. The cluster had not been planned in detail by anyone. Li had funded the institutional preconditions, Chang had built the anchor tenant, and the gravitational physics of the foundry model had pulled the rest of the ecosystem in. By 1995 the science park employed about fifty thousand people and accounted for several percent of Taiwan’s GDP. By 2000 TSMC alone, by then a publicly traded company on the Taiwan Stock Exchange and listed in New York via American Depositary Receipts, was the largest taxpayer in Taiwan.

Li lived to see most of it. Sun Yun-suan, who had pulled Chang into the original 1982 conversation, had suffered a debilitating stroke in 1984 and never returned to active politics; he died in 2006, having watched from the sidelines as the industry he had midwifed in the 1974 breakfast meeting at the Taipei soy-milk shop became the centerpiece of his country’s economy. Li died in May 2001, after a long retirement during which he was treated by every Taiwanese administration that succeeded him as a kind of philosophical advisor on technology policy. Chang, at the inaugural Li Kwoh-ting Award ceremony in November 2023, would say plainly that without K.T. Li, TSMC would not have existed. He meant the sentence as more than a courtesy. Li had been the official who had identified the chip industry as a strategic priority before any rational economic calculation supported the choice, who had recruited a foreign-trained executive on the assumption that an outsider’s instincts were precisely what the project required, and who had then sheltered the founder from the bureaucratic and political weather of a one-party state long enough for the company to clear the runway. The chip industry of the modern world has produced very few examples of state policy that worked at this scale on this timeline. Li’s was one of them, and it had worked because Li had understood early what most of his peers in his own government and in his American counterparts had not yet understood: that semiconductors had become a kind of sovereign infrastructure, and that countries that did not have a chip industry were going to spend the rest of the century being countries that did not have a chip industry.

The deeper consequence of the founding, the one that even Chang and Li in 1987 did not see clearly, was structural rather than economic. A pure-play foundry was not just a different kind of company. It was a different kind of industry. The vertically integrated chipmakers of the 1980s had treated manufacturing as a private competence, captive to a particular firm, indistinguishable from that firm’s design strategy. By unbundling the two, Chang had made manufacturing a service, available on commercial terms to anyone who could draft a layout. The consequences of that unbundling would not finish playing out for thirty years. They included, in roughly the order they would matter: the explosion of fabless chip startups in the 1990s; the migration of the world’s most advanced logic manufacturing into a single industrial park in Hsinchu; the slow transformation of the United States from the country that designed chips and made them into the country that designed chips and bought them; the construction of a Taiwanese economic indispensability that would by the 2010s be described, half in flattery and half in alarm, as a silicon shield; and, in the end, the geopolitical question that would dominate the back half of the next century, which was what happened to a world whose digital infrastructure depended on the political stability of an island ninety miles off the coast of an authoritarian state that claimed it.

In Hsinchu in 1987, the question did not yet exist. The fab Chang had built was running 2-micron CMOS for whoever would buy it. The third floor of the ITRI building was full of process engineers learning to read American customer specifications written in idioms unfamiliar to them. The first foundry of its kind in the world had opened. The customers it had been built for had not yet shown up. The bet had been made, and would not be settled in any final form for years. Li, when he visited the fab with Chang in the spring after the founding, looked at the wafers on a tray and asked, in his careful English, whether they were any good. Chang said they were getting better. Li said that was what mattered. Taiwan, in February 1987, had the beginnings of a semiconductor industry. In a quiet way, the rest of the world had a problem it would take several decades to recognize.